Low-Dimensional HRTF Characterization Through Common-Pole/Zero Modeling and Statistical Component Analysis

Abstract

Head-related transfer functions (HRTFs) provide a mathematical representation of the acoustic filtering imposed by the human head, torso, and external ears and constitute a fundamental component of three-dimensional spatial audio reproduction. However, the high dimensionality of measured HRTF datasets creates substantial challenges for storage, interpolation, transmission, and real-time rendering. This paper develops a low-dimensional HRTF characterization approach that integrates common-pole/zero modeling with statistical component analysis. The proposed methodology first represents individual HRTFs through a shared acoustic pole/zero structure and subsequently transforms the resulting parameter space into a compact statistical representation. The approach is theoretically motivated by established common-acoustical-pole/zero models and finite- and infinite-impulse-response HRTF representations. The proposed framework separates relatively stable acoustic characteristics from location-dependent variation, thereby reducing the number of parameters required to describe a spatial HRTF dataset. A methodological analysis demonstrates how common-pole/zero modeling can provide physically interpretable spectral parameters while component analysis can reduce redundancy among those parameters. The study further examines reconstruction, dimensionality, computational efficiency, and potential deployment in real-time spatial-audio systems. The resulting framework offers a structured compromise between acoustic fidelity, parameter compactness, and computational complexity. Its principal significance lies in combining parametric acoustic modeling with statistical dimensionality reduction rather than treating either technique independently.